PARAMETER ESTIMATION AND TRACKING OF MANEUVERING TARGETS WITH THE PARAMETRIC AVALANCHE

WE PROPOSE TO SOLVE TWO IMPORTANT PROBLEMS IN THE TRACKING OF MANEUVERING TARGETS IN DYNAMIC IMAGERY. THE FIRST IS THE COMPUTATIONAL COMPLEXITY OF THE STOCHASTIC FILTER, E.G., THE KALMAN FILTER; AND THE SECOND IS THE USE OF TARGET SHAPE AND ORIENTATION TO ENHANCE TRACKING PERFORMANCE THROUGH VIOLENT MANEUVERS. OUR PROPOSAL INVOLVES A CASCADED CONFIGURATION OF OUR PROPRIETARY PARAMETRIC AVALANCHE NEUROCOMPUTING ARCHITECTURE, WHICH IS A MASSIVELY PARALLEL IMPLEMENTATION OF THE CONTINUOUS TIME BAYESIAN ESTIMATOR BASED ON THE INNOVATIONS METHOD OF T. KAILATH AND THE PRINCIPLE OF QUANTUM NEURODYNAMICS OF R. DAWES. THE PHASE I PROJECT WILL RESULT IN COMPUTER SIMULATIONS OF SMALL BUT NONTRIVIAL NETWORKS OF PARAMETRIC AVALANCHE MODULES TO TEST AN IMPORTANT THEORETICAL CAPABILITY OF SUCH NETWORKS, NAMELY, TO FACTOR IMPUT SIGNALS INTO STOCHASTICALLY ORTHOGONAL SUBSPACE REPRESENTATIONS. THIS WILL IN TURN DEMONSTRATE BOTH THE COMPUTATIONAL EFFICIENCY OF THE METHOD AND ITS ABILITY TO USE ABSTRACT INFORMATION, SUCH AS TARGET SHAPE, FOR THE IMPROVEMENT OF CONCRETE ESTIMATES, SUCH AS THE TARGET TRAJECTORY.